# The Safe Asset Cliff: Demographic Downgrade Risk and Collateral Scarcity

**Brian Peters**

March 2026

## Abstract

Population aging simultaneously increases demand for safe sovereign assets and undermines the fiscal foundations of safe issuers. This paper develops a probabilistic model of sovereign downgrade risk linked to demographic aging. Using a 31-country panel of S&P sovereign ratings (1990-2024), I estimate ordered logit and PanelGLS models that identify a critical OADR threshold: above 20% old-age dependency, each additional percentage point of aging reduces sovereign ratings by 8.6 points on the 21-point S&P scale (β = -17.5, p < 0.003 on the spline term). Monte Carlo simulations project that the number of safe issuers (AA- or above) falls from 24 to a median of 13.5 by mid-century, with the safe supply ratio declining from 0.89 to 0.77. The expenditure-revenue gap emerges as the primary fiscal channel through which aging precipitates downgrades, while r-g compression provides a partial but diminishing buffer. These findings formalize the self-reinforcing paradox at the heart of safe asset scarcity: the demographic forces that elevate demand for safe assets are the same forces that erode the supply.

## 1. Introduction

The global financial system rests on the assumption that a substantial pool of sovereign assets carries negligible credit risk. Pension funds, insurance companies, central bank reserves, and collateral frameworks all depend on the availability of "safe" government debt. Yet the same demographic transition that is driving unprecedented demand for safe assets — population aging that swells retirement savings and pension obligations — simultaneously threatens the fiscal sustainability of the sovereign issuers that produce them.

This paper bridges two strands of the demographic-capital flow literature. Peters (2026b) documents the expenditure-revenue asymmetry in aging economies: a 3.3:1 ratio whereby each percentage point increase in old-age dependency raises government expenditure by 33 basis points of GDP while revenue rises only 10 basis points, generating persistent fiscal pressure on safe issuers. Peters (2026c) establishes that aging compresses r-g for safe issuers by 26.5 basis points per percentage point of OADR increase, providing a partial buffer against debt sustainability concerns. What remains unquantified is whether these opposing forces ultimately preserve or destroy safe issuer status — and when the tipping point arrives.

The challenge is fundamentally small-N. The historical record contains only 20 sovereign downgrade events among the 31 countries that have held investment-grade ratings since 1990, and only 7 instances where a country lost safe status (fell below AA-). This is too few for a standard Cox proportional hazard model to deliver reliable inference. I address this through four complementary approaches: (1) pooled logit for any-downgrade prediction, which provides 20 events; (2) complementary log-log (cloglog) for the rare loss-of-safe event; (3) ordered logit over the full five-category rating distribution; and (4) PanelGLS with the continuous 21-point rating scale, which uses all rating variation and yields the highest statistical power.

The key finding is a demographic nonlinearity. Below 20% OADR, aging is weakly associated with higher ratings — consistent with demographic demand supporting bond markets and compressing yields. Above 20%, the relationship reverses sharply: each additional percentage point of OADR reduces ratings by 8.6 notches on the spline term (β = -17.5, p < 0.003). This threshold is precisely the range where the expenditure-revenue asymmetry begins to overwhelm the r-g compression buffer.

The headline deliverable is a Monte Carlo stress test. Drawing 1,000 coefficient vectors from the estimated variance-covariance matrix and projecting UN WPP medium-variant demographics through 2054, I find that the median number of safe issuers falls from 24 to 13.5, with the 10th percentile showing as few as 8 remaining. France, Finland, Hong Kong, the Czech Republic, and Taiwan face the highest downgrade probabilities. The safe supply ratio — aggregate safe debt relative to global GDP — declines from 0.89 to 0.77 at the median, with a 10th-percentile scenario of 0.13 representing near-total safe asset scarcity.

These projections have direct implications for collateral frameworks, central bank reserve management, and the design of safe asset pooling mechanisms such as the European Safe Bonds (ESBies) proposal.

## 2. Literature

The safe asset literature identifies a growing tension between supply and demand. Caballero, Farhi, and Gourinchas (2017) develop a framework where safe asset scarcity drives down equilibrium interest rates and creates a safety trap analogous to the liquidity trap. Gorton and Ordoñez (2022) model safe assets as information-insensitive securities whose status depends on maintaining low perceived credit risk — a status that can be lost abruptly when fiscal conditions deteriorate.

He, Krishnamurthy, and Milbradt (2019) provide the theoretical foundation for the downgrade cascade mechanism. In their model, a sovereign's safe status depends on the stock of debt relative to a fiscal capacity threshold; demographic pressure that widens expenditure-revenue gaps moves issuers toward this threshold. Critically, the loss of safe status by one major issuer shifts demand to remaining safe issuers, potentially triggering second-round fiscal pressure.

The empirical sovereign rating literature identifies key determinants. Cantor and Packer (1996) establish that per capita income, GDP growth, inflation, external debt, and economic development explain the cross-section of sovereign ratings. Afonso, Gomes, and Rother (2011) extend this to a panel framework, finding that fiscal balance, government debt, and the current account are significant predictors of rating changes. Kopecky and Taylor (2020) link demographic aging to sovereign risk through the fiscal channel. None of these studies, however, estimates the demographic threshold at which safe status becomes threatened or projects the aggregate safe supply implications.

Peters (2026a) establishes the demographic polynomial framework linking population age structure to capital flows. Peters (2026b) quantifies the fiscal dominance channel through which aging undermines sovereign fiscal positions. Peters (2026c) documents the safe asset demand-supply paradox at the aggregate level. This paper closes the loop by modeling the individual-country downgrade process and aggregating to safe supply projections.

## 3. Data and Methodology

### 3.1 Data

The analysis uses a balanced panel of 31 countries with S&P sovereign credit ratings spanning 1990-2024 (1,021 country-year observations). Rating data is compiled from S&P historical rating actions as documented in Kose, Kurlat, Ohnsorge, and Sugawara (World Bank compilations). Ratings are mapped to a 21-point numeric scale (AAA = 21 through SD/D = 0), with "safe issuer" status defined as AA- (18) or above.

Demographic variables come from the UN World Population Prospects 2024 revision. The key variable is the old-age dependency ratio (OADR), defined as the population aged 65+ relative to the working-age population (15-64). The demographic polynomial terms Z₁, Z₂, Z₃ follow the Koomen specification used in Peters (2026a). Forward demographics ($OADR_{t+10}$, $OADR_{t+20}$) are from UN WPP medium-variant projections.

Fiscal variables — government debt/GDP, expenditure/GDP, revenue/GDP, primary balance, structural balance — are from the IMF World Economic Outlook. The expenditure-revenue gap (exp_rev_gap = G/Y - T/Y) captures the fiscal stress channel identified in Peters (2026b). The interest rate-growth differential (r-g) is constructed from 10-year government bond yields and nominal GDP growth. Macroeconomic controls include real GDP growth, inflation, and the Chinn-Ito capital account openness index (KAOPEN).

### 3.2 Event Variables

I construct four event variables from the rating panel:

- **downgrade_any**: equals 1 if rating_numeric decreased from the prior year (20 events)
- **downgrade_notch**: magnitude of downgrade (0 if none; range 1-8 notches)
- **lost_safe**: equals 1 if rating fell below AA- from safe status in the prior year (7 events)
- **lost_aaa**: equals 1 if rating fell below AAA from AAA in the prior year (9 events)

Additionally, I construct a five-category ordered variable: AAA (4), AA+ (3), AA (2), AA- (1), Below AA- (0).

### 3.3 Estimation Strategy

**Logit models.** For binary downgrade prediction, I estimate pooled logit:

$$P(downgrade_it = 1 | X_it) = Λ(X_it β)$$

where Λ is the logistic CDF. Standard errors are from the Hessian of the log-likelihood. I report marginal effects at means, McFadden pseudo-R², AUC, and Brier scores.

**Complementary log-log.** For the rare loss-of-safe event (7 events in 884 at-risk observations), I use the cloglog link function:

$$P(lost_safe_it = 1 | X_it) = 1 - exp(-exp(X_it β))$$

which is appropriate when the event probability is small and asymmetric.

**Ordered logit.** To use all rating variation, I estimate ordered logit over the five-category rating variable:

$$P(Y_it ≤ j | X_it) = Λ(α_j - X_it β)$$

with K-1 cutpoints estimated jointly with the coefficient vector.

**PanelGLS.** The continuous rating model uses PanelGLS with iterative Cochrane-Orcutt AR(1) correction:

$$rating_it = α + X_it β + u_it,  u_it = ρ u_{i,t-1} + ε_it$$

This maximizes power by using the full 21-point rating variation. The OADR spline specification allows for a structural break at 20%:

$$rating_it = α + β₁ OADR_it + β₂ max(0, OADR_it - 0.20) + γ'X_it + u_it$$

### 3.4 Monte Carlo Stress Test

To convert econometric estimates into policy-relevant projections, I conduct a Monte Carlo simulation:

1. Draw 1,000 coefficient vectors from the multivariate normal distribution defined by the estimated coefficients and variance-covariance matrix.
2. For each draw, project rating_numeric for all current safe issuers through 2054 using UN WPP medium-variant demographic projections and historical fiscal trend extrapolation.
3. Apply the AA- threshold to classify each country as safe or not-safe in each simulation.
4. Aggregate: safe supply = sum of debt/GDP times GDP for countries still rated safe.
5. Report median, 10th/90th percentile safe supply ratios by year.

## 4. Descriptive Results

### 4.1 Rating Transitions

Table 1 presents the annual rating transition matrix across the five categories. The dominant pattern is stability: AAA-rated country-years transition to AAA 97% of the time. However, conditional on leaving AAA, the transition tends to be to AA+ (gradual) rather than to lower categories. Below-AA- ratings show substantial persistence (96% remain below), consistent with the "rating trap" identified in the corporate bond literature.

### 4.2 Demographics and Ratings

Table 2 documents the demographic-fiscal profile by rating category. AAA-rated observations have a mean OADR of 0.20 and government debt of 43% of GDP. Below-AA- observations have a mean OADR of 0.18 but substantially higher debt (67% of GDP) and wider expenditure-revenue gaps. The non-monotonicity in OADR across categories reflects the dual role of aging: moderate aging is associated with high-income status and strong institutions (supporting high ratings), while extreme aging undermines fiscal sustainability.

Figure 2 displays the OADR-rating scatter with downgrade events highlighted. The downgrade events cluster in two regions: (1) moderate OADR with fiscal crises (Ireland, Spain, Korea in the 1990s-2010s), and (2) high OADR with gradual fiscal deterioration (Japan, Italy).

### 4.3 Safe Issuer Decline

Figure 1 shows the number of safe issuers over time. The count peaked at approximately 22 in the early 2000s and has remained around 24 since the post-GFC recovery of Ireland and Korea's upward migration. However, the composition has shifted: several countries that were AAA in 2000 (France, UK, Austria, Finland) are now AA or AA+.

## 5. Econometric Results

### 5.1 Downgrade Prediction

Table 4 presents the logit models for any-downgrade prediction. In the demographics-only specification, the Z-polynomial captures modest predictive power (Pseudo-R² = 0.03, AUC = 0.67). Adding OADR level improves discrimination (AUC = 0.64), but the substantial improvement comes from fiscal variables: the full model with OADR, debt/GDP, expenditure-revenue gap, primary balance, and GDP growth achieves Pseudo-R² = 0.21 and AUC = 0.87. Higher debt/GDP (β = 0.017, p < 0.05) and lower GDP growth (β = -0.23, p < 0.001) are the strongest predictors.

The predetermined specification using $OADR_{t+10}$ achieves the highest discriminatory power (AUC = 0.89), suggesting that forward-looking demographic information improves downgrade prediction beyond contemporaneous controls.

### 5.2 Loss of Safe Status

Table 5 reports cloglog models for the loss-of-safe event, restricted to country-years with lagged rating at or above AA-. With only 7 events in 884 at-risk observations, power is limited. Nevertheless, the expenditure-revenue gap emerges as a significant predictor (β = 0.25, p < 0.001, in the full specification), consistent with the fiscal dominance channel. The full model achieves AUC = 0.88 despite the low event count.

These results carry an explicit low-power caveat. The confidence intervals are wide, and the point estimates should be interpreted as indicative rather than precise.

### 5.3 Ordered Logit

Table 6 presents the ordered logit results, which exploit the full five-category rating variation. The OADR coefficient is strongly positive (β = 7.57, p < 0.001), indicating that higher OADR is associated with higher-category ratings — reflecting the correlation between aging, income, and institutional quality. Debt/GDP (β = -0.014, p < 0.001), the expenditure-revenue gap (β = -0.051, p < 0.001), and inflation (β = -0.128, p < 0.001) all push toward lower rating categories. This specification captures the competing forces: aging is associated with both higher demand for safe assets (supporting ratings) and fiscal deterioration (undermining them).

### 5.4 Continuous Rating Model

Table 7 presents the PanelGLS results for the continuous rating scale. The OADR spline specification (Model 3) is the key result:

$$β_OADR = 8.86 (SE = 4.01, p < 0.05),  β_spline(20%) = -17.49 (SE = 5.79, p < 0.003)$$

Below 20% OADR, aging is associated with higher ratings (consistent with the income-institutions channel). Above 20%, the net effect of aging on ratings is 8.86 - 17.49 = -8.63 per percentage point of OADR — economically and statistically significant. The crossover point where the marginal effect turns negative is at 0.20 + 8.86/17.49 = 0.506, or approximately 51% OADR — but the spline specification implies that the damage begins accruing immediately above 20%.

The dynamic specification (Model 6) reveals strong rating persistence (ρ_lag = 0.95, p < 0.001), with the expenditure-revenue gap (β = -0.012, p < 0.001) and 5-year debt change (β = -0.005, p < 0.001) as significant conditional predictors. This confirms that fiscal deterioration drives rating changes even after controlling for rating inertia.

### 5.5 Interaction Tests

Table 8 tests whether demographic and fiscal variables interact in driving ratings. Two interactions are statistically significant:

**OADR × Debt/GDP** (β = -0.048, p = 0.042 in PanelGLS): At higher OADR, additional debt has a more negative effect on ratings. This is the He-Krishnamurthy-Milbradt channel — demographic demand stops protecting safe status when debt levels are high.

**r-g × OADR** (β = 0.114, p = 0.016 in PanelGLS): The r-g compression from aging partially buffers ratings from demographic pressure. However, this buffer diminishes as OADR rises further and fiscal costs accelerate.

The OADR × exp-rev gap interaction is not statistically significant, suggesting that the fiscal channel operates primarily through debt accumulation rather than through contemporaneous flow imbalances.

### 5.6 Safe Supply Stress Test

Table 9 presents country-by-country rating projections, and Table 10 reports the aggregate safe supply distributions. The Monte Carlo simulation produces the following headline results:

The median safe supply ratio declines from 0.89 in 2024 to 0.77 by 2054. The 10th percentile falls to 0.13, representing a scenario of near-total safe asset scarcity. The expected number of safe issuers falls from 24 to 13.5 at the median.

The most vulnerable current safe issuers are:

- **Finland**: projected rating falls to 16.5 by 2054 (below A+), driven by rapid aging and fiscal expansion
- **France**: projected rating reaches 17.4 by 2054 (A+), with the expenditure-revenue gap widening
- **Hong Kong**: projected rating drops to 17.2 by 2054, reflecting aging acceleration
- **Czech Republic**: falls to 17.2, moving below safe threshold by the 2040s
- **Taiwan**: declines to 17.8, at the margin of safe status

The most resilient safe issuers are Luxembourg (stable at 19.9, young demographics and fiscal surplus), Singapore (19.4), and the Gulf states (stable demographics and low debt).

Under the aging acceleration scenario (UN WPP high-variant demographics), the median number of safe issuers falls to 11, with Finland, France, Hong Kong, Czech Republic, Taiwan, Netherlands, Denmark, Canada, and Korea all losing safe status by 2054.

## 6. Discussion

### 6.1 The Self-Reinforcing Paradox

These results formalize a self-reinforcing paradox. As populations age:

1. **Demand rises**: pension funds, insurance companies, and aging savers increase their allocation to safe assets.
2. **Supply falls**: the same aging that drives demand erodes the fiscal foundations of safe issuers through the 3.3:1 expenditure-revenue asymmetry.
3. **Scarcity intensifies**: fewer safe issuers must absorb greater demand, supporting their bond prices and compressing r-g.
4. **The buffer weakens**: once OADR exceeds 20%, the fiscal costs of aging overwhelm the r-g compression benefit, and ratings begin to decline.

The OADR spline result — a threshold at 20% — is consistent with the fiscal dominance finding that the OADR coefficient on r-g for safe issuers is -26.5*** (Peters, 2026b). Below 20%, demographic demand compresses yields sufficiently to offset fiscal pressure. Above 20%, the cumulative fiscal burden exceeds the r-g buffer, and downgrade risk rises sharply.

### 6.2 Cascade Risk

The He-Krishnamurthy-Milbradt framework suggests that safe asset markets are subject to cascade dynamics. If a major issuer (e.g., France) loses safe status, demand shifts to remaining issuers, potentially supporting their ratings through yield compression. However, the interaction results show that the OADR × Debt/GDP interaction is negative — high-debt issuers benefit less from demand shifts. This creates the possibility of a second-round cascade where the demand shift to remaining issuers with already-high debt actually increases their downgrade probability.

### 6.3 Policy Implications

The projections suggest three policy-relevant conclusions:

1. **Sovereign safe bonds**: pooling mechanisms like the European Safe Bonds (ESBies) proposal become more urgent as individual-country safe supply contracts. The senior tranche of a diversified sovereign bond pool can maintain safe status even as individual components deteriorate.

2. **Collateral frameworks**: central bank collateral frameworks that rely on ratings thresholds will face shrinking eligible pools. The ECB's current framework, for example, uses a rating threshold that several Eurozone members may breach within the projection horizon.

3. **Fiscal consolidation**: the OADR spline result implies that countries approaching the 20% threshold have a window of opportunity to build fiscal buffers before the adverse dynamics accelerate. The interaction tests suggest that lower debt/GDP at the critical OADR threshold substantially reduces downgrade risk.

## 7. Connection to Companion Papers

This paper is part of a series examining demographic influences on international capital flows and asset markets:

- Peters (2026a) establishes the demographic polynomial framework and estimates the baseline current account model across 175 countries.
- Peters (2026b) quantifies the fiscal dominance channel: the 3.3:1 expenditure-revenue asymmetry and the heterogeneous r-g effect (-26.5, p < 0.001 for safe issuers). The expenditure-revenue gap variable used in this paper's downgrade prediction models directly operationalizes that finding.
- Peters (2026c) documents the safe asset demand-supply paradox at the aggregate level and introduces the safe gap ratio measure. The rating projections in this paper provide the micro-foundation for the safe supply contraction documented in that paper.

Together, these papers establish a complete chain: demographics → fiscal pressure → r-g compression → rating deterioration → safe supply contraction → collateral scarcity.

### Geoeconomic Power Implications

An and Huber (2026) demonstrate that geoeconomic power in global FX funding markets --- measured as a country's capacity to redirect cross-border capital through policy actions --- is time-varying and responds to fiscal events. Notably, US geoeconomic power declined visibly following the Tax Cuts and Jobs Act and CBO deficit upward revisions. This finding extends our safe asset cliff projections into a new dimension: as aging-driven fiscal pressure erodes safe-issuer creditworthiness, the affected countries simultaneously lose geoeconomic leverage in FX funding markets. Japan, the only An-Huber country that has already lost safe-issuer status (rating 17, below the AA- threshold), holds just 7% of aggregate geoeconomic power --- far below what its GDP share would predict. The OADR dispersion across the eight currencies in their sample nearly quadruples from 3.2pp (2000) to 12.1pp (2050), suggesting that the competitive structure of global FX funding markets will become increasingly unstable as demographic divergence among major currency issuers widens. Our Monte Carlo projection of safe issuer attrition (24 $\to$ 13.5 by 2054) implies not merely a collateral shortage but a reorganization of the institutional architecture through which global capital is intermediated.

## 8. Robustness

The OADR spline result is robust across multiple tests (Table 11):

**Leave-one-country-out jackknife**: The OADR coefficient ranges from 7.8 to 10.1 across the 31 jackknife samples (full sample: 8.9, SD: 0.45). No single country drives the result.

**Pre/post-GFC split**: The spline coefficient is significant in both subperiods (pre-GFC: -17.6**, post-GFC: -16.3**), though the base OADR coefficient is only significant pre-GFC.

**OECD subsample**: The effect is concentrated in OECD countries (OADR: 20.2**, spline: -29.5***), consistent with the rating variation occurring primarily among advanced economies.

**Placebo test**: Permutation of demographic variables across 200 shuffles yields a mean OADR coefficient of -0.01 with SD 0.35. The true coefficient (8.86) is 25 standard deviations from the permutation mean (p < 0.001).

**Pseudo-OOS backtest (Table 12)**: We estimate the OADR spline model on 1990--2010 data and project forward to 2024 using actual demographics. The model achieves RMSE of 1.76 versus 1.98 for a naive no-change benchmark (relative RMSE = 0.89), and correctly classifies 91% of countries' safe-issuer status (21/23). However, the Monte Carlo intervals are overconfident: 80% coverage is only 48%, and the model completely misses the eurozone-crisis downgrades of Spain (AAA $\to$ A) and Italy (AA+ $\to$ BBB). These two countries are assigned P(safe) $\geq$ 0.97 by the model, revealing that the demographic framework captures the slow-moving *level* of sovereign creditworthiness but not the tail-risk of crisis-driven downgrades. The model's strength --- and its limitation --- is structural: it identifies which countries are demographically vulnerable to safe-status erosion, not when a sudden stop or fiscal crisis will trigger the actual downgrade. For policy purposes, this means the Monte Carlo projections in Section 5.6 should be interpreted as demographic vulnerability rankings rather than as calibrated probability forecasts.

## 9. Conclusion

This paper establishes that demographic aging creates a quantifiable cliff in sovereign creditworthiness. The OADR threshold at 20% marks the transition from a regime where aging supports safe asset demand to one where fiscal deterioration dominates. Monte Carlo projections indicate that the number of safe issuers will decline from 24 to approximately 14 by mid-century, with the safe supply ratio contracting by 14% at the median and potentially by 85% in the adverse tail.

These findings underscore the urgency of institutional innovation in safe asset production. As the demographic transition narrows the base of sovereign safe issuers, the financial system's dependence on a shrinking pool of safe collateral creates systemic fragility. Sovereign safe bond pooling mechanisms, reformed collateral frameworks, and proactive fiscal consolidation in countries approaching the OADR threshold represent the policy toolkit for managing this transition.

## References

Afonso, A., Gomes, P., & Rother, P. (2011). Short- and long-run determinants of sovereign debt credit ratings. *International Journal of Finance & Economics*, 16(1), 1-15.

Caballero, R. J., Farhi, E., & Gourinchas, P. O. (2017). The safe assets shortage conundrum. *Journal of Economic Perspectives*, 31(3), 29-46.

Cantor, R., & Packer, F. (1996). Determinants and impact of sovereign credit ratings. *Economic Policy Review*, 2(2), 37-54.

Gorton, G., & Ordoñez, G. (2022). The supply and demand for safe assets. *Journal of Monetary Economics*, 125, 1-18.

He, Z., Krishnamurthy, A., & Milbradt, K. (2019). A model of safe asset determination. *American Economic Review*, 109(4), 1230-1262.

Kopecky, J., & Taylor, A. M. (2020). The murder-suicide of the rentier: Population aging and the risk premium. NBER Working Paper 26943.

Peters, B. (2026a). Demographics and multilateral capital flows: A demographic polynomial approach. Working Paper.

Peters, B. (2026b). Fiscal dominance and the demographic transition: Expenditure-revenue asymmetry in aging economies. Working Paper.

Peters, B. (2026c). The safe asset paradox: Demographic demand and sovereign supply constraints. Working Paper.
